Hybrid Statistical Models for Forecasting of Rice Production in Karnataka State, India

Abstract

Rice (Oryza sativa L.) is a vital staple crop underpinning food security and agricultural livelihoods in Karnataka and across India. The current study used both linear and nonlinear time series techniques to anticipate Rice production in Karnataka state from 1962–1963 to 2021–2022. First, the Autoregressive Integrated Moving Average (ARIMA) model was used, and the best fit was chosen using diagnostic metrics like the Coefficient of Determination (R2), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE). The ARIMA (2,1,2) model was found to be the best suitable of the linear models. Advanced machine learning techniques like Time Delay Neural Network (TDNN), Nonlinear Support Vector Regression (NLSVR) and their hybrid combinations with ARIMA (ARIMA–TDNN and ARIMA–NLSVR) were used to identify potential nonlinear patterns in the data. The adoption of hybrid models was justified by the BDS test on ARIMA residuals, which verified the existence of nonlinearity. RMSE, MAE and MAPE were used to assess the model performance for the nonlinear and hybrid techniques. The most recent three years of data were used for testing, while 57 years of data were used for training in the adopted training and testing framework. In terms of forecast accuracy, the ARIMA (2,1,2)–TDNN (3–7–1) hybrid model outperformed the other models. Forecasts up to 2027–28 was also produced using the model, estimating that Karnataka would produce 4432.68 thousand tons of Rice. These findings imply that, in comparison to individual models, hybrid models that include linear and nonlinear features offer more accurate and consistent forecasts for agricultural production series.